Report comment

Anon @2:03 - the problem is (setting aside the opposition that Westlaw, Lexis Nexis and the like would have to your proposal) that this isn't how LLM work.

As ChatGPT admitted to Mr Gough, they work by identifying patterns and using these to predict sequences that allow them to produce human-like text. What they don't currently do is analyse in the way a human would. Even with access to legal databases, they wouldn't be able to identify how a case applies to an original set of facts, balance dissenting judgements, distinguish a case from a precedent etc. What they would do is produce very erudite and forceful arguments, using the correct case citations, but with a significant risk of still getting the application of those cases wrong - this would be even more dangerous than the current situation where fake citations are at least pretty easy to spot.